Loading...
Loading...
Found 45 Skills
This skill provides comprehensive guidance for adapting Wan-series video generation models (Wan2.1/Wan2.2) from NVIDIA CUDA to Huawei Ascend NPU. It should be used when performing NPU migration of DiT-based video diffusion models, including device layer adaptation, operator replacement, distributed parallelism refactoring, attention optimization, VAE parallelization, and model quantization. This skill covers 9 major adaptation domains derived from real-world Wan2.2 CUDA-to-Ascend porting experience.
PyTorch-based TAO image classification. Supports a wide range of backbones (FAN, EfficientNet, ResNet, etc.) with distillation and quantization for deployment. Use when training, evaluating, distilling, quantizing, exporting, or running inference for a TAO image-classification (PyT) model. Trigger phrases include "train image classifier", "TAO classification", "ResNet/EfficientNet/FAN backbone classifier", "classification-pyt".
Reduce LLM API and infrastructure costs through model selection, prompt caching, batching, caching, quantization, and self-hosting strategies. Track spend by team and model, set budgets, and implement cost-aware routing.
Qdrant vector database: collections, points, payload filtering, indexing, quantization, snapshots, and Docker/Kubernetes deployment.
Export a promoted fine-tuned model in the right deployment format — merged safetensors, LoRA-only, GGUF with imatrix, or FP8. Use after a checkpoint passes promotion, when choosing a quantization format for a target device, or when an exported model fails its smoke test.
Debug silent corruption when a MAX model loads, compiles, serves, and generates tokens but output disagrees with a reference implementation. Use whenever parity debugging stalls on scalar taps, the model returns gibberish or wrong greedy tokens, logit cosine is high but argmax differs, or generation is coherent then diverges — during an architecture port, a quantization bring-up, a multi-GPU conversion, or after a MAX upgrade. Triggers on "parity failure", "silent corruption", "logits match but tokens diverge", "top-1 mismatch", "greedy divergence", and "model serves but generates garbage". Not for crashes on load or pre-serve scaffolding (use import-model). Mandates reference-vs-MAX tensor-dump comparators first, verify fixes numerically before recompiling, and serve-vs-pipeline bisect when dumps match but text diverges.
Apply quantization to reduce memory by 4-32x. Enable HNSW indexing for 150x faster search. Configure caching strategies and implement batch operations. Use when optimizing memory usage, improving search speed, or scaling to millions of vectors. Deploy these optimizations to achieve 12,500x performance gains.
Use when integrating Foundation Models framework, implementing on-device AI with Apple Intelligence, building tool-calling AI features, working with guided generation schemas, converting models with Core ML and coremltools, or running open-source LLMs on Apple Silicon. Covers Foundation Models (LanguageModelSession, @Generable, @Guide, SystemLanguageModel, structured output, tool calling), Core ML (coremltools, model conversion, quantization, palettization, pruning, Neural Engine, MLTensor), MLX Swift (transformer inference, unified memory), and llama.cpp (GGUF, cross-platform LLM).
Memory-efficient fine-tuning with 4-bit quantization and LoRA adapters. Use when fine-tuning large models (7B+) on consumer GPUs, when VRAM is limited, or when standard LoRA still exceeds memory. Builds on the lora skill.
Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.
Convert HuggingFace transformer models to ONNX format for browser inference with Transformers.js and WebGPU. Use when given a HuggingFace model link to convert to ONNX, when setting up optimum-cli for ONNX export, when quantizing models (fp16, q8, q4) for web deployment, when configuring Transformers.js with WebGPU acceleration, or when troubleshooting ONNX conversion errors. Triggers on mentions of ONNX conversion, Transformers.js, WebGPU inference, optimum export, model quantization for browser, or running ML models in the browser.
Use the Geti application end to end through its REST API — the project → dataset → annotate → train → deploy pipeline served by the FastAPI backend in `application/backend/`. Use when a user (not a contributor) wants to create a project, upload media, add annotations, launch a training or quantization job, track job status, configure a source → model → sink inference pipeline, and enable live inference. Covers the `/api/...` endpoints and the async job model, not backend code changes.